BP神经网络PID在费托反应器汽包压力控制中的应用

Application of the BP neural network PID in Fischer-Tropsch reactor steam drum pressure control

费托合成反应是天然气制油(GTL)或煤制油(CTL)生产技术的核心。在费托合成反应器换热系统中汽包压力的控制具有大惯性、大滞后、强非线性等特点,参数固定不变的传统PID控制器难以进行精确的压力控制。为了解决这一问题,设计了费托合成反应器换热系统汽包压力BP神经网络自整定PID控制器。在MATLAB中实现了BP神经网络自整定PID程序的编写,并通过MATLAB与组态王(KINGVIEW)的动态数据交换(DDE),方便的实现了上位机与控制系统中其它硬件的通信。在此基础上进行了具有大惯性、大滞后汽包压力控制的实验,实验结果表明,在汽包压力的定值控制和阶跃变化控制中BP神经网络自整定PID控制器的静态偏差维持在-0.03 MPa^+0.01 MPa,最大误差为1.5%,并且对系统的非线性变化具有一定的适应性,其控制效果明显优于参数固定的传统PID控制器。

Fischer-Tropsch（FT） synthesis is an important stage in gas to liquids（GTL） or coal to liquids（CTL） technology. The steam drum pressure control of heat exchange system for Fischer-Tropsch synthesis faces many challenges, such as big inertia, large lagging, high nonlinearities, etc. The traditional Proportional-Integral-Differential（PID） controller with fixed parameters is hard to obtain satisfied pressure control performances. In order to solve this problem, the Backword Propagation（BP） Neural Network self-tuning PID control algorithm was successfully implemented in MATLAB software. Data communication between the controller and other hardwares was realized through KINGVIEW by dynamic data exchange method（DDE）.Finally the BP neural network PID was applied to the Fischer-Tropsch reactor steam drum pressure control experiments. Experimental results show that, in set point control and step change control of the steam drum pressure, static deviation of BP neural network self-tuning PID controller was-0.03 MPa ~ ＋0.01 MPa, and the maximum error was 1.5 %; moreover, it shows certain adaptation to the nonlinear change of the reaction system, revealing it was superior to traditional PID controller.